Can I use claude-context without an OpenAI key (OpenRouter, Mistral, or local LM Studio)?

6 answers

• Best answerAgentArmor teamTeam3d ago

  Status: explained — no dedicated providers for Mistral, LM Studio, or Bedrock exist yet, but any OpenAI-compatible embedding endpoint works today. The project does not use an LLM, so you cannot point it at Claude Code or OpenRouter LLM endpoints; it needs an embedding model. OpenRouter itself only added embedding support in November 2025, so a commenter noted you could swap in an OpenRouter embedding model via the base URL and key.

  One commenter confirmed a working setup with Mistral's free embedding model codestral-embed-2505 and a local Milvus instance using this mcp.json:

  jsonCopy to clipboard

  {
    "mcpServers": {
      "claude-context": {
        "type": "stdio",
        "command": "npx",
        "args": ["@zilliz/claude-context-mcp@latest"],
        "env": {
          "EMBEDDING_PROVIDER": "OpenAI",
          "EMBEDDING_MODEL": "codestral-embed-2505",
          "OPENAI_API_KEY": "YOUR_API_KEY",
          "OPENAI_BASE_URL": "https://api.mistral.ai/v1",
          "MILVUS_ADDRESS": "127.0.0.1:19530"
        }
      }
    }
  }
  

  They also provided the docker-compose stack for Milvus (etcd, MinIO, Milvus standalone v2.6.2) to run locally. For local servers, a commenter suggested setting the base URL to the OpenAI-compatible endpoint (LM Studio, llama.cpp) and noted that another commenter confirmed LM Studio ignores the API key value entirely, so any placeholder works. AWS Bedrock had no confirmed setup.

  Security note: keep the real API key out of version control if you commit mcp.json.

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What a GitHub user tried on GitHub (Outcome unclear)

  Asked whether the OpenAI key is mandatory and whether an OpenRouter key or Claude Code itself could be used instead.

  Source: https://github.com/zilliztech/claude-context/issues/81 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @edxeth tried on GitHub (Worked)

  Configured EMBEDDING_PROVIDER as "OpenAI", set EMBEDDING_MODEL to codestral-embed-2505, OPENAI_API_KEY to a Mistral key, OPENAI_BASE_URL to https://api.mistral.ai/v1 (opens external site) ↗, and used a local Milvus at 127.0.0.1:19530. Confirmed working in Claude Code.

  Source: https://github.com/zilliztech/claude-context/issues/81#issuecomment-3313465983 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @zc277584121 tried on GitHub (Partly worked)

  A maintainer explained the project only requires an embedding model, and suggested replacing the base URL and API key variables if the provider supports embeddings. This is guidance; later comments noted OpenRouter added embedding support in November 2025.

  Source: https://github.com/zilliztech/claude-context/issues/81#issuecomment-3133192228 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @prom3theu5 tried on GitHub (Partly worked)

  Reported that with LM Studio the API key can be any value because it is ignored upstream, but did not describe a full working configuration.

  Source: https://github.com/zilliztech/claude-context/issues/81#issuecomment-3462489636 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @rujialiu tried on GitHub (Partly worked)

  Suggested exposing the embedding endpoint via an OPENAI_API_BASE_URI-like setting for local OpenAI-compatible servers; a later comment pointed out OPENAI_BASE_URL already exists for this.

  Source: https://github.com/zilliztech/claude-context/issues/81#issuecomment-3159987324 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport